Firehose
Filtered to tagged “Machine learning” · clear filters
Browse: People · Companies · Papers · Podcasts · Hacker News · Deep dives
Browse by tag
Claude Fable 5.2, the latest model from the company, has been released with significant power but also some concerns about its vulnerability to errors and its ability to compete with other models like Claude Opus 5.5. Jensen Huang, CEO of Nvidia, has been accused of having an affair with a Chinese spy, leading to a potential scandal involving the company's board of directors and search for an interim CEO. A new report on AI R&D acceleration suggests that full recursive self-improvement and superintelligence could arrive within a year, with some experts predicting a "material risk of AI takeover in the next 6-12 months." AI summary
The Gemini Decisions API is a fast, calibrated multimodal decision-making tool that can evaluate up to 1,000 yes/no, choice, and rubric-scoring questions in a single pass across text, images, audio, video, and PDFs. It achieves 150-250 ms end-to-end latency at $0.10 per 1M input tokens with zero output tokens billed. AI summary
Apple is developing a new Mac app called Harbor that enables users to run local AI models on their Macs without sending data to the cloud, utilizing Thunderbolt 5 connectivity to pool unified memory across multiple devices. Harbor automatically detects and configures devices, allowing users to download and run open-weight models without manual setup. The app is currently in private preview under the Claris brand and is expected to launch publicly in the near future. AI summary
Mux is the video API that translates audio and answers questions about your video. Get $50 in free credits: https://mux.com/fireship Anthropic just dropped Fable and Mythos 5.2 in the middle of a rocky IPO roadshow. Let's dive into what the…
Claude Fable 5.2 (Max, Default Fallback) scores 65 on the Artificial Analysis Intelligence Index, placing it above average among comparable models, but is notably slow at 44.4 minutes per Intelligence Index task and expensive at $26.54 per task. It has a high output token count of 275M, which is among the most verbose models in its class. AI summary
Anthropic has reported that their AI research is dramatically accelerating, with capabilities progress doubling in just a few months. This acceleration is likely due to AI automating AI research, rather than increased compute or other inputs. Researchers must be sped up by a factor of 2-10x to achieve this level of acceleration, which could lead to extreme superintelligence within a year. AI summary
This episode features John Jumper, co-creator of AlphaFold, discussing the breakthrough in protein structure prediction that won the 2024 Nobel Prize in Chemistry. Jumper explains the technical innovations behind AlphaFold, its impact on bi…